v0.2.0

Changes since v0.1.0

• b285aa4 chore: release v0.2.0
• d99e5b0 fix(ui): avoid main-window localization injection
• da28093 fix(annotations): classify unlabeled summaries
• e9d73a3 fix(ui): preserve item-list context menu
• 52a169b fix(annotations): persist classification tags
• 8e39991 fix(ui): isolate sidebar keyboard handling
• 1b32ca5 fix(classification): apply configured label colors
• 15f1394 Merge branch 'main' of https://github.com/00sapo/zotero-skimming
• bb366dc fix(ui): restore sidebar after actions
• f54a8dd refactor(classification): reset summary label config namespace
• c9d87ed feat(ui): move skimming controls to sidebar
• cc37b9d refactor(classification): remove legacy zero-shot classifier
• 4484b84 feat(classification): map labels from summary sentences
• 0aa5b67 Update README.md
• c252834 Create LICENSE
• 74ad0ee Update README.md

Full history

v0.1.0

Changes since 0.0.3

• 1727776 chore: release v0.1.0
• 930f319 fix(zero-shot): repair silent classification failure and add enable flag
• 34bac49 feat: replace binary coverage flag with dynamic matrix-based MMR coverage
• dede873 updated readme
• c27a554 feat: add zero-shot classifier module with TOML label config
• f7fb602 chore: remove keywordSections from scoring-config, fix release asset path
• 5d3c47f refactor: replace tagDefinitions with zeroShotConfig in config and NLP
• 8e5df8f feat: add context-aware help text below Source toggle
• e0c62e0 feat: add remote embeddings support with default text-embedding-3-small
• 685f39c feat: reorganize settings into NLP analysis / API sections
• e3a1046 refactor: use local prompt format identically for remote LLM
• 317c331 feat: change defaults to local/ratio-10, add model override fields
• 3c61151 fix: register pref defaults, force summary sentence count, add logging
• e6f5d14 feat: add --only-build flag to release script
• e482eaf chore: remove dead code and unused exports
• 46e39ea refactor: dead code removed
• 0ffcc51 dev: release --only-build flag
• 7873476 chore: release v0.0.4
• 686599b fix: suppress thinking tokens in local Ollama summarization
• 3f16527 chore: full release script with gh release creation
• b0652ea chore: remove old build.sh, superseded by scripts/release.sh
• 790fd49 chore: add release script and fix update_url
• 44ab09b refactor: rename add-on IDs and names to zotero-skimming
• 7324f41 feat: map-reduce sentence chunks with N selector, disable thinking
• f9f3199 feat: sentence-based summarization with compression ratio
• 52f8194 updated models
• 44d5389 feat: load model identifiers from model-identifiers.json
• b3a0acf fix: use model-identifiers.json model paths
• 30350d6 fix: use real ollama model names
• 08a5417 feat: test ollama button, remove model upload
• f57b33a fix: replace fetch with XMLHttpRequest for Ollama requests
• ef141f5 fix: shorten validation message
• c72f7cc fix: generic error message instead of misleading density message
• e7a5fd3 fix: specific error for non-absolute ollama command
• c9aa3b8 fix: require absolute ollama command path
• 1fc06f6 fix: use XMLHttpRequest for model downloads
• b0e26ae fix: back to Qwen2.5 with independent chunk MR
• 7fe8177 fix: switch summarization to mT5
• 3977000 feat: rename test api to test summary
• c5f8467 feat: embed summary sentences and reward coverage in MMR
• 6218e00 fix: readded sentence classification via re-ranking
• 09979b1 fix: upload ollama gguf blobs
• 4d9da82 fix: restore local model cache keys
• 79180b2 feat: refine local map reduce summaries
• 6878b91 feat: run local inference in worker
• fb1e6bc fix: simplify model loading status
• 495cdb3 fix: unify summarization context window
• bd10d90 feat: add local qwen summarization
• 2102c4b docs: add similar plugins section
• f09fa16 feat: simplify summary ranking input

Full history

0.0.3

Full Changelog: 0.0.2...0.0.3

0.0.2

Full Changelog: 0.0.1...0.0.2

0.0.1

First release.

• Summarization
• Embedding retrieval
• Maximal Marginal Relevance to maximize summary coverage
• Encourage short sentences
• TF-IDF when embeddings lack
• Sentence classification

Sub-sentences are not handled correctly and pose a computational problem, so they are not in this release. It's likely easier to ask the LLM the list of sentences than to evaluate them via embeddings.